Clientelism in the Public Sector

Transcription

1 Public Disclosure Authorized Policy Research Working Paper 8439 WPS8439 Public Disclosure Authorized Public Disclosure Authorized Clientelism in the Public Sector Why Public Service Reforms Fail and What to Do about It Tessa Bold Ezequiel Molina Abla Safir Public Disclosure Authorized Education Global Practice & Social Protection and Jobs Global Practice May 2018

2 Policy Research Working Paper 8439 Abstract In many developing countries (and beyond), public sector workers are not just simply implementers of policies designed by the politicians in charge of supervising them so called agents and principals, respectively. Public sector workers can have the power to influence whether politicians are elected, thereby influencing whether policies to improve service delivery are adopted and how they are implemented, if at all. This has implications for the quality of public services: if the main purpose of the relationship between politicians and public servants is not to deliver quality public services, but rather to share rents accruing from public office, then service delivery outcomes are likely to be poor. This paper reviews the consequences of such clientelism for improving service delivery, and examines efforts to break from this bad equilibrium, at the local and national levels. This paper is a product of the Education Global Practice and the Social Protection and Jobs Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at research. The authors may be contacted at tessa.bold@iies.su.se, emolina@worldbank.org, and asafir@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team

3 Clientelism in the Public Sector: Why Public Service Reforms Fail and What to Do about It Tessa Bold Ezequiel Molina Abla Safir 1 JEL classification codes: H0, I0 Keywords: Public Service Delivery, Public Sector Governance 1 Bold: Institute for International Economic Studies, Stockholm University; Molina: Education Global Practice, World Bank; Safir: Social Protection and Jobs Global Practice, World Bank. This paper is a background paper for the World Development Report 2017 Governance and the Law. We thank Kathleen Beegle for very useful comments. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.

4 1. Introduction The World Development Report 2017: Governance and the Law (World Bank 2017) highlights the intimate connection between the effectiveness of policy reforms and governance. The Report argues that power asymmetries play an important role in ensuring that policy reforms are credible and overcome collective action problems; with one particular manifestation being clientelism. 2 Further, it notes that in order to expand the set of implementable policies, there is need to change the policy arena by: (a) changing incentives; (b) reshaping preferences; and (c) increasing the contestability of the decision-making process. In this background paper, we will focus on how power structures affect incentives for policy reforms and ultimately outcomes in the context of public service delivery. Here, we have a particular power structure in mind, namely when public servants themselves hold power. In many developing countries (and beyond), public servants are not just the agents tasked with delivering services by the principal (the clients of the service, usually represented by politicians), they are also elites, in the sense that they can have direct influence on policy design and implementation. This has implications for the quality of public services: if the main purpose of the relationship between principal and agent is not to deliver quality public services, but rather to share rents accruing from public office, then service delivery outcomes are likely to be poor. Breaking such an equilibrium may be difficult and successful policy reform needs to take these kinds of power constraints into consideration. We first make the case that public servants aside from delivering services may capture rents in a multitude of ways: through the allocation of jobs, through above-market wages, and through low performance on the job, including with absenteeism or moonlighting. We then review the research that tries to explain why it is possible for public servants to capture rents and under what circumstances clientelistic equilibria, in which politicians transfer rents to 2 Clientelism is a political strategy characterized by giving material goods in return for electoral support (Stokes 2009). Clientelism can be viewed as a two-party encounter between a politician and a voter (Hicken 2011). It is, however, often organized in networks, which can be based on districts or regions. As a result, a central part of clientelism s organizational structure is an intermediary or a broker, whose role is to mobilize a network of local voters in exchange for financial payment or patronage jobs (WDR 2017, based on Wantchekon 2016). 2

5 public servants in exchange for political support, are likely to arise. This research also suggests why public sector reform may be so difficult: if rent-sharing arises as part of a tacit agreement between politicians and public servants in which rents are transferred in exchange for political support, then any reform that tries to make public servants more accountable and reduce their rents will likely be seen as reneging on such an agreement and be met with opposition. In the second part of the paper, we review research that has focused on making public servants more accountable. Here, we first note that successful accountability reforms may face somewhat of a trade-off between two of the constraints to policy implementation discussed in the Report, namely the power and the capacity constraints. We then focus on two strands of literature, one that experimentally evaluates policies to increase accountability of public servants locally, often through the involvement of the main stakeholders, the users of public services. This, mainly experimental literature, usually takes the political power constraints as given, and highlights the importance of information and the identity of those monitoring the public servant. We discuss to what extent such local reforms can be successful. Second, we review a set of more descriptive papers from the political science literature that discuss successful and unsuccessful attempts to change the political equilibrium in which public services are delivered and attempt to analyze which factors make reform success more likely. 2. How Do Public Sector Workers Capture Rents? In this section, we examine the ways in which rents can be shared in the public sector: first, through the allocation of jobs and pay that exceeds private sector options, and second, through shirking on the job. Jobs and Pay Instead of delivering high-quality services (or at least the services allowed by their capacity), the public sector may be trapped in a clientelistic equilibrium in which jobs are awarded in exchange for political support and represent an inefficient mechanism of transferring resources to supporters. 3

6 There is some evidence that the allocation of public sector jobs may be politically motivated. Alesina et al. (2001) estimate that half of the public wage bill in the South of Italy can be interpreted as a subsidy from the richer North to the poorer South, both because of an overinflated size of the public sector and through a public wage premium compared with outside alternatives. Calvo and Murillo (2004) argue that the attractiveness of exchanging public sector jobs for political support depends on the characteristics of a party s clientele. The poorer and less educated a party s supporters, the more attractive it is to offer public sector jobs in exchange for political support. This is the case because lower education implies a lower wage premium needs to be offered and patronage can therefore be extended to more people. They find evidence for this in Argentina: provinces governed by the Peronist Party, whose supporters are traditionally less educated and poorer, provide a higher public wage premium to low-skilled workers than provinces governed by the Union Cívica Radical, a party of the urban middle classes, and employ 21 percent more public employees. This holds even after controlling for income measures. In the other direction, the authors show that voters indeed reward politicians for this form of patronage: increases in public employment have a significant and positive effect on the probability of electing the Peronist Party. Closely related to the allocation of public sector jobs as a way to bestow patronage is the issue of wage setting in the public sector. After all, a public sector job can only be used as a reward if the public sector wage is higher than the outside options, for the same level of effort. There is indeed substantial evidence of a `raw public-private sector wage gap in most low and some middle-income countries. In Zambia, the premium stands at percent (SkytNielsen and Rosholm 2001) and in Tanzania at 51 percent (Lindauer and Sabot 1983). In a number of countries in Latin America, it ranges from 40 percent in Chile to 111 percent in Colombia (Mizala et al. 2011), and in Pakistan it is estimated to be roughly 50 percent (Aslam and Kingdon 2009; Hyder and Reilly 2005). Finan et al. (2015) show for a large cross-section of countries that there is a public sector wage premium and that this tends to be higher in lowincome countries. Nonetheless, for this to be evidence of rent-sharing, one needs to be able to show that for a given worker the wage in the public sector is higher than her outside option. Most of the papers in this literature attempt to estimate this `causal pay gap by controlling for worker 4

7 characteristics. However, even if one could perfectly control for selection, the differences in wages might still be a consequence of efficiency wages being paid to elicit higher effort. A notable exception is Barton et al. (2016) who use a regression discontinuity design that exploits the Kenyan government s decision to hire roughly 18,000 new teachers in In each constituency, candidates were ranked according to a simple algorithm that rewarded grades in secondary school and teacher training as well as length since graduation from teacher training. Comparing successful and unsuccessful candidates close to the cut-off, the authors find a return of around 50 percent to employment as a public sector teacher (compared to the outside option). Given that the large majority of unsuccessful candidates also worked as teachers on short-term contracts with lower pay but in identical jobs and the fact that experimental studies show that these types of teachers are no less (if not more) effective than permanently employed public sector teachers (Duflo et al. 2015; Muralidharan and Sundararaman 2013), the authors can rule out (or at least cast doubt on) compensating differentials and efficiency wage explanations of this wage differential. Instead, rent-sharing between politicians and public sector workers seems the most likely explanation. Lower Effort Once on the job, public sector workers can capture rents by exerting low effort, in its most blatant form not showing up for work. Bold et al. (2016) show that low effort, in particular in the form of absenteeism, is rife among teachers in public primary schools in Sub-Saharan Africa. Analyzing nationally representative data from 7 countries (Senegal, Tanzania, Kenya, Uganda, Togo, Mozambique and Nigeria), they find that during a random unannounced spot check, almost half the teachers (45 percent) are not in the classroom teaching. In Mozambique, almost two-thirds (61 percent) are not in the classroom teaching. As a result, students are missing out on half the scheduled teaching time each day and the authors find that this is negatively related to learning. Across five low- and middle-income countries (Bangladesh, Ecuador, Indonesia, Peru and Uganda), Chaudhury et al. (2006) find that 35 percent of health workers were absent from the facility during an unannounced visit. Bold et al. (2012) report that clinicians in Senegal spend an average of 39 minutes per day counseling patients, while in Tanzania the number is even lower at just half an hour. 5

8 Capture through low effort can also manifest itself more subtly, when public sector workers do not accumulate the knowledge and skills necessary to provide quality public services or do not apply them when required. For example, Das et al. (2008) find that doctors in Delhi spend on average 3.8 minutes with a patient, ask 3.2 questions and perform just one examination. In Tanzania, 28 percent of doctors spend three minutes per patient and on average perform no physical examination with the average patient. Using vignettes for patients representing with malaria, Bold et al. (2012) find that only just over half of Tanzanian health providers can correctly diagnose the disease in a country where malaria is endemic. Similarly, over two-thirds of Tanzanian and Senegalese clinicians cannot diagnose diarrhea with severe dehydration correctly. A powerful study observing the same doctors in their public sector jobs and in their private practice indicates that the low effort is a problem of incentives, although capacity is also a problem. In a representative sample of rural areas of Madhya Pradesh, doctors in public facilities spend on average 2.4 minutes with a patient and complete only 16 percent of a checklist of examination items and questions on medical history. The same doctors perform better when they are in the private sector indicating the importance of incentives. However, virtually no doctors conducted all the examinations indicated when a child had diarrhea. Patients are much more likely to receive an unnecessary treatment than a correct treatment. Only 3 percent gave a correct treatment (Das et al. forthcoming). In a more descriptive setting, Bold et. al (2016) find that absence is significantly lower in private schools than in public ones across Sub-Saharan Africa. That public sector workers capture resources intended for the beneficiaries of the service has also been demonstrated in experimental studies. In an experiment conducted by Duflo, Dupas, and Kremer (2015) with public sector teachers in government schools in Western Kenya, schools were randomly allocated an additional teacher, employed on a short-term contract, to be placed in grade 1. In some schools, school management committees were also trained to monitor this teacher. The authors find that in the absence of this training, absenteeism among the existing teachers (all on permanent contracts) increased by 16 percent if an additional contract teacher was placed in the school. Of course, none of this is to say that this way of capturing resources is intrinsic to public sector workers. Rather, we argue that weak accountability might be intrinsic to the sector if 6

9 public sector jobs are a reward for political support. That is not to say, that motivated public servants do not exist and do not bemoan this equilibrium. For example, a health assistant in Ethiopia noted that absenteeism in public facilities was a rational response to the fact that moonlighting in the much better paid private sector was not regulated in any way. We observe [absenteeism] we cannot lie about it. Not respecting working hours is something we got used to and this has created a problem for the patients. During tea time, for example, people go to work in the private sector. Even the small amount of time we have in the facility we do not use effectively (see Lindelow and Serneels qualitative study of Ethiopian nurses, 2006). Similarly, poor infrastructure may beget low effort if the two are complements. As noted by an Ethiopian health assistant in the same study: There is inadequate supply of material [in the public sector]. It is not possible to work efficiently and meet expectations. If you try to work with what is available, your service will be sub-standard, and eventually you might be forced to drop your ethics. Blocking Reforms That Threaten Rents If public sector jobs are a mechanism to distribute rents to political supporters, then reforming public service employment to be more accountable reduces rents and would potentially be interpreted as the patron reneging on his part of the bargain. This may be one (among other) reasons why teacher unions tend to be opposed to accountability reforms in the public sector especially if these might imply salary reductions for some teachers. However, while teacher unions tend to be opposed to pay reform, this is by no means the case for the average teacher (as shown for the US by Ballou and Podgursky, 1993) partly because not all of them are part of the clientelistic equilibrium (for example in Kenya about 20 percent of public primary school teachers are employed by Parent-Teacher Associations rather than the government at a fraction of the pay of civil service teachers) or even within a clientelistic equilibrium, effective and motivated teachers stand to gain from such reforms. Mirroring Ballou and Podgursky s findings in the US, Muralidharan and Sundararaman (2011) show that in their sample of government teachers in Andra Pradesh over 80 percent of teachers are in favor of merit pay and that exposure to actual incentive pay far from demoralizing teachers actually increases teachers support of it. This split in teachers interest became visible for example in the US, when teachers in Washington, DC, were asked to vote on 7

10 performance-pay reviews: the vast majority of young teachers voted for the change, while the rest did not (see Bruns and Luque, 2014). We will show below how this can be exploited in successful accountability reforms. Does It Matter? There is much evidence that links rent-seeking on the job, for example through low effort and shirking, to worse outcomes, such as low learning achievements (Duflo et al. 2015; Bold et al. 2016). Rent-sharing in the form of diversion of financial resources - corruption also affects learning outcomes. In Brazil, students test scores in mathematics and Portuguese are higher when corruption is lower in the municipalities where the schools are located (Ferraz, Finan, and Moreira 2012). There is some evidence that links political connections, government jobs, and poor service provision or poor outcomes. In Pakistan, more politically connected doctors are more likely to be absent and less likely to be sanctioned for their absences (Callen et al. 2014). In India, despite rules that are supposed to protect the independence of the India Administrative Service, politicians frequently intervene in the careers of bureaucrats. In particular, bureaucrats of the same caste as the incoming politicians party base secure important positions. Suggestive evidence indicates that in states with less interference from politicians in the transfers of civil servants, poverty rates declined faster (Iyer and Mani 2012). However, there is need for more research to link rent-sharing and clientelistic relationships that originate at higher levels of the service delivery chain with outcomes at the provider level. 3. Why Can Public Sector Workers Capture Rents? When the Agent and the Principal Are Reversed Why do clientelistic equilibria arise and why is it public sector jobs (rather than direct income transfers) that are used to bestow patronage? Robinson and Verdier (2013) argue that it is the consequence of a two-sided commitment problem between voters and politicians. Voters cannot commit to vote in a particular way and public sector jobs are a credible, selective and reversible method of redistribution, which ties the continuation utility of a voter to the political 8

11 success of a politician. On the flipside, the authors argue that politicians find it easier to commit to allocating public sector jobs to supporters rather than income transfers, because the former generate a rent to the politician whereas the latter do not; politicians can exercise control on public sector workers through influencing their careers. For this reason, it is in the interest of public sector workers to support politicians since the latter can help them in their careers. Other models have postulated that it may be easier for politicians to justify financing supporters through awarding public sector jobs, which can be claimed to have some socially desirable outcomes, rather than through income transfers (Alesina, Reza, and Easterly 2000). More generally, different models of clientelism tend to show that it is more likely to arise the lower the return to private sector activities, as that makes patronage through public sector jobs easier, the higher the stakes to staying in power and the more unequal and fractionalized the society. The relationship between inequality and clientelism arises because inequality makes collective action more difficult, which means it may be easier to look for targeted support. Consistent with this, Alesina, Reza and Easterly (2000) find that public employment in US cities is larger the more unequal the city s income distribution is and the more ethnically fragmented it is. This may be a reason why developing countries, with their lower capacity and lower cooperation to begin with, including in relation with low human capital and fractionalization, are more likely to have clientelistic policies. In such contexts, it may be difficult for politicians to credibly commit to broad public policies that are difficult to monitor, and it is easier to commit to targeted benefits because they can more credibly deliver them. A first form of clientelism can be to obtain the votes of citizens in exchange for income transfers (Khemani et al. 2016). However, some groups or actors may wield more power and support than others. This is a second form of clientelism, whereby politicians may become dependent on the support of certain groups, including public sector workers. In such settlings, citizens do not expect broad or long-term policies from politicians (because they receive some targeted benefits), while providers extract rents (because they play a role for politicians reelection). In the extreme, the principal-agent relationship between policy maker (the principal) and public servant (the agent), as described in the World Development Report 2004: Making Services Work for People (World Bank 2003), may be inversed in a clientelistic equilibrium. Carpenter (2001) and Moe (2006) (cited in Gailmard, 2014) note this in the context of 9

12 bureaucrats and legislators, where the former may assemble electoral coalitions to influence the identity and ideology of who their legislative principals are in the first place. In the context of public sector employees, Mexico provides a striking example of the ability of public sector employees in this case the union representatives of teachers to capture government, both historically and to the present day. As detailed in Bruns and Luque (2014), in 2012, eight of 31 of Mexico s state-level education ministers were union appointees and many other key positions were held by the union. At the legislative level, the union has throughout its history maintained close links with Mexico s long-standing ruling party, PRI. This has led to an asymmetry in power between organized teachers interests and government in Mexico, [that] has been one of the most striking in the world. In such settings, workers may consider that their principal, responsible for monitoring them, is the union rather than the politicians representing the users. While not quite to the same degree, similar examples of union influencetaking abound in Latin America, with profound implications for the ability to reform public services that we will explore below. Another example is the Indian state of Bihar. Kingdon and Muzammil (2009) document, in a case study, that a large number of deputies of the state assembly are teachers. Through their political role and ability to shape policies, teachers have improved in-work benefits; there has however been less progress for broader improvements in the education system. While one might expect that as countries grow, the clientelistic relationships may weaken, there is reason to think that the very existence of these relationships may hinder development. For example, Robinson and Verdier (2013) show in their theoretical model that the existence of patronage may give incentives to politicians to restrict beneficial public investment since that would increase the outside option of voters (and therefore the cost of patronage) and because it reduces the rents that accrue to the patron and therefore makes patronage through public sector jobs less credible. But restricting public investment is, of course, overall inefficient. 10

13 4. Policies to Make Public Servants More Accountable: Can Reforms That Only Increase Monitoring or Address Local Power Constraints Succeed? A Trade-Off Between Capacity and Power Constraints? First off, it is important to note that not all elites, be they public sector workers or other types of elites who can have direct influence on policy design or implementation, capture rents. In addition, capture where it exists may not necessarily be very large. Lastly, elites may have leadership skills and knowledge that the general population does not. Therefore, the desirability to address elite capture depends on the magnitude of capture compared with the efficiency costs associated with circumventing elites. Typically, unions have played an important role, historically in aggregating the preferences of workers and helping both structure their demands and negotiate with employers, the most successful example being the solidaristic bargaining in Northern Europe (Moene and Wallerstein 2006). Just like more generally in the case of elite capture, whether public sector workers capture rents is context-specific. Moreover, circumventing elites in an effort to avoid capture can also have costs in terms of the quality of service delivery. In the broader case of elite capture, there is indeed variation in its extent and it depends on local context and project design. Khwaja (2009) attempts to estimate the benefits of local leadership involvement in development projects. Based on data from 132 community maintained infrastructure projects in 99 communities in Northern Pakistan, he estimates how project maintenance is related with a number of community and project features. Khwaja (2009) finds that both community and project characteristics matter for project maintenance. In particular, social heterogeneity affects maintenance negatively, while there is a U-shaped relationship between land inequality and maintenance. Importantly, he finds that presence of a leader is significantly positively related with better maintenance and that this effect increases substantially with project complexity. In line with the theory, community participation is beneficial in non-technical decisions, but has a negative effect on maintenance for technical ones. 11

14 Khwaja s paper is also of interest because it shows that variation in project characteristics explains at least half the variation in maintenance. In other words, even in communities where community characteristics predict collective action problems, good project design can increase maintenance. This is an important finding because it shows that even when wholesale reform that fully addresses the political economy equilibrium is not possible, some progress can be made by changing conditions within a given equilibrium. While this does not directly apply to public sector workers, it is a reminder of the fact that whether elite capture is a problem is context-specific and depends on the cost and benefits of elite participation in a program as argued by Khwaja (2009) the costs to efficient program implementation of circumventing elites can be significant. In the context of public service delivery, Rasul and Rogger (forthcoming) make an important contribution estimating the impact of variation in management practices on the quality of public services in Nigeria. Based on a data set of 4,700 projects (essentially construction projects, such as boreholes, buildings, dams and roads) across 63 organizations, the authors link management practices in different agencies to the likelihood that a project will be completed (including measurements of the quality of the project). While their data are not experimental, they argue that their estimates of the impact of management practices on public services can be considered as the coefficients in a production function because they can condition on many project and implementing agency characteristics and because they compare the same project type (i.e. training, infrastructure, etc.) across different ministries. Their findings are striking: for the same project type, they find that the percentage of projects never started varies from 11 to 95 percent, those fully completed varies from 3 to 89 percent and projects rated satisfactory from 25 to 100 percent. Similarly, they find that against the backdrop of a civil service which is widely considered corrupt and in which 31 percent of the projects considered were reported to have some corruption issues a one standard deviation increase in management practices (specifically giving bureaucrats more autonomy) is associated with a 17 percent increase in project completion rates. Moreover, there is no interaction between the level of corruption associated with a project and management practices. In other words, even for a given corruption level, a change in management practices can have positive effects. Conversely, curtailing autonomy (an often advocated counter corruption measure) is associated with lower project completion. 12

15 Is It Possible to Improve Service Delivery with Local Interventions? Given the rent-seeking constraints that we have described, service delivery is not poor by accident; rather it is a symptom of the underlying institutional environment. And the argument here has been that this institutional environment is often put in place to share rents between politicians and public sector workers. The question then becomes of course how we go from an institutional environment characterized by clientelism and rent-sharing to one in which quality services are delivered. Reforms are not easy, since clientelism operates at a high level in what Acemoglu (2010) describes as the seesaw effect: When reforms threaten powerful insiders, large scale implementation may provoke political economy reactions, creating an endogenous policy response that counteracts the objectives of reform. Local interventions to increase cooperation for monitoring One could think that local power constraints play a role in capture: For instance, that teachers and doctors (or in short, endline providers) capture some of the rents from a program. One way to strengthen service delivery might therefore be to strengthen collective action locally and this has been the motivation behind a host of community participation interventions, primarily meant to (informally) influence norms and collective action and thereby local decision making. In an experiment in Western Kenya, Duflo, Dupas, and Kremer (2015) provide both evidence of local rent-sharing in action and how to mitigate it. The authors study the effects of a contract teacher program managed by a local NGO, in which half of 140 government primary schools were randomly selected to receive funds to hire an additional teacher on a short-term basis. In addition, half of the schools that received a teacher also received school-based management training, in which parents received a short course in how to hire and manage the teacher. In treatment schools, the contract teacher was placed in grade 1, which was split, and students were randomly assigned to the contract teacher or the existing civil service teacher. 13

16 When comparing test scores in treatment and control schools, the authors find a significant difference of standard deviation only in those schools that received both an additional teacher and school-based management training. The overall treatment affect (averaging across schools with and without school-based management training) is 0.14, just shy of significant. The pure contract effect (i.e. comparing to a civil service teacher in the treatment school) of contract teachers is 0.25 of a standard deviation. Importantly, they find no effect of the class size reduction on test scores in the classrooms of existing teachers and an increase in teacher absence of 15.7 percentage points among regular teachers in treatment schools. Together, this suggests that the increase in resources implied by the halving of pupil-teacher ratios was almost fully captured by the existing teacher. Similarly, rents were captured at the hiring stage. In schools that received an additional teacher but no training, 34 percent of contract teachers hired were relatives of existing regular teachers and this had a negative effect on test scores. The authors show that where parents were trained and collective action therefore presumably strengthened capture was significantly reduced. In those schools that also received training, the decrease in attendance by the regular teacher was only 8.3 percentage points, i.e. half of what it was in schools that received a teacher, but no training. Similarly, in schools where parents were trained, the share of relatives among contract teachers was 15 percentage points lower, just shy of significant. In these schools, there was less of a negative effect from hiring relatives. This suggests that training parents on the school management committee both reduced local capture and mitigated its effect where it did happen. Importantly, the authors also provide suggestive evidence that there may have been long-run effects of the training initiative on the capacity for collective action: schools with management training for the parents were almost twice as likely to raise funds to keep the contract teacher after the program ended. The role of information to improve monitoring at the local level: Contrasting evidence Björkman and Svensson (2009) provide evidence on an intervention designed to strengthen local accountability and community-based monitoring (CBM) in the primary health care sector in Uganda. The intervention consisted of a series of community meetings facilitated by an NGO. In the first meeting, communities were presented with baseline information on the quality of services in their community collected in report cards. Based on the report cards and 14

17 further discussions, communities were then asked to draw up action plans on how to improve service quality. This was followed by a meeting of the community and the health care providers in which the findings from the report card were discussed and a shared action plan was agreed on. This intervention was remarkably successful in improving both health services and outcomes in the participating communities. Utilization for outpatient services increased by 20 percent and treatment practices, waiting time, examination procedures and absenteeism all improved significantly. Most importantly, the authors found a significant increase in the weight of infants and a reduction of under-5 mortality by one-third in the treatment villages. Björkman, De Walque and Svensson (2015) revisit the original communities for a longterm follow-up and also compare these long-term effects to an alternative, cheaper CBM program that `replicated the community meetings but without disseminating report card information on benchmark services. The authors find that the effects on utilization, practices and health outcomes persist in the long-term follow-up. However, they find that the CBM program without information dissemination had no effects on either service delivery or health outcomes. This indicates that without relevant information, it is difficult to strengthen local monitoring. This interpretation is strengthened by the fact that in the communities where report card information was provided, the action plans focused on local problems that could be directly addressed by the health provider and the community. In contrast, without information provision, the action plans focused on issues that could only be resolved by third parties, such as additional financial support. The authors conclude that information makes it easier to identify which actions of the health care provider contribute to low-quality services and therefore help to turn the focus to issues under the control of the community and the service provider. Another promising result on the impact of information comes from Pakistan, where a randomized experiment that provided information to parents on the performance of private and public schools increased test scores, decreased private school fees, and increased primary enrollment (Andrabi, Das, and Khwaja 2015). 15

18 The type of information conveyed to the community is extremely important, however. In particular, it has to be actionable for the community mobilization to have `bite (Fox 2015). This is illustrated by an intervention evaluated by Banerjee et al. (2010) who examine a threefold intervention to increase community participation in school monitoring. In the first treatment, communities were mobilized to be made aware of and strengthen the local village education committees tasked with monitoring schools in a series of meetings run by a large NGO. In the second treatment, the NGO also disseminated information on the state of learning and specifically reading during the community meetings. In the third treatment, in addition to training parents to assess reading ability (as in intervention 2), communities were also trained in how to improve shortfalls in reading: select volunteers were recruited from the village to undertake reading camps. In other words, one might argue that this additional component made the information `actionable. While the first two interventions failed to increase either parents awareness of the village education council or their engagement with it and therefore had no effect on performance, the third intervention had a large effect on reading scores. As the authors note, this suggests that, [ ], information combined with the offer of a direct channel of action can result in collective action and improved outcomes. In another experiment in Kenya, providing information on children s performance in schools and how parents could take action did not increase their participation in monitoring (Lieberman, Posner, and Tsai 2014). The authors emphasize the many implicit assumptions that link the provision of information to improving services; two important ones in this case are whether parents think that monitoring services is their responsibility and whether they can do anything about it. In particular, information in local interventions may improve outcomes only when constraints related to asymmetries in bargaining power are alleviated. While solid evidence is lacking, some information-related interventions may have been more successful because they did not face constraints related to a lack of bargaining power among actors. For instance, two information campaigns on the performance of local schools in northern India had contrasting outcomes. One campaign that was supported by the Department of Education was successful in improving outcomes (Pandey, Goyal, and Sundararaman 2009). Another campaign that was supported by an NGO did not improve the performance of schools (Banerjee et al. 2010). One interpretation is that in the case of the campaign supported by the Department of Education, constraints related to the top-down delivery of services had already been alleviated (Khemani 16

19 and others forthcoming). When such constraints are still present, information may have little or no impact. When local monitoring is only effective if credible and powerful When trying to constrain the rent-sharing of public servants, whether those tasked with it have actual power is crucial. For instance, in the study of Banerjee et al. (2010) reviewed above, the composition of the village education committee consisting of the head teacher, the elected head of village government, and three parents chosen by local officials, implied that the village education committee could not be considered an independent monitor (Fox 2015). In a study set in Indonesia, Pradhan et al. (2014) examine two different ways of strengthening school committees tasked with monitoring schools. In the first treatment, they affect the composition of school committees by randomly assigning them to be constituted by democratic election. In the second treatment, school committees, which are considered relatively powerless, were given `bite by being linked to the village council, a more powerful body, through a series of facilitated joint planning meetings. The authors find that the first treatment, democratic elections of school committee members, increases awareness but has no effect on learning. On the other hand, linking school and village committees increases test scores by 0.17 standard deviation, and linkage plus election increases them by 0.23 after two years, pointing to potential complementarities between the composition of the school committee and synergizing it with a more powerful external body. The mechanism by which linkage and election increased test scores was through increases in community contributions (0.14 and 0.13 standard deviation). Importantly, increases in contributions were linked to concrete joint initiatives between school committee and village councils, such as hiring contract teachers or the institution of a village study hour, suggesting that the intervention truly improved collective action in the treated villages. 5. Policies to Make Public Servants More Accountable: The Long Route of Reforms Limits to the Effectiveness of Local Reforms 17

20 However, there are also limits to how much local control can achieve. This is partly the case - as noted by Björkman et al (2015) - because there are important components to the quality of service delivery that are not determined locally. For example, most service providers are public servants whose pay and contract structure are determined centrally, not locally. Secondly, even though beneficiaries of the service may in principal have the incentive to hold service providers to account, they may not be able to do so in practice, even when encouraged to do so through participatory interventions. As noted by Muralidharan and Sundararaman (2013) in their evaluation of contract teachers in India, the fact that elites (in this case regular teachers) capture rents through higher wages is not just a potential symptom of a clientelistic equilibrium, it may also perpetuate it by making it even harder for beneficiaries to exert local control. Citing Kingdon (2011), they note that these high salaries may contribute to the large social distance between communities and teachers and make it more difficult for parents to hold teachers to account. This issue is illustrated in work by Cilliers et al. (2013) who evaluate an absence monitoring intervention in Ugandan primary schools. Specifically, they randomly assign 180 Ugandan primary schools to one of four treatments: (i) monitoring of teachers absence by the head teacher, (ii) monitoring of teachers absence by parents, (iii/iv) teacher attendance is rewarded with a financial bonus/no bonus; and a control group. Their first finding is that both head teachers and parents underreport substantially: the actual absence rate measured in independent spot checks is 14 percentage points larger than the reports from parents and head teachers. Counter to expectations, however, they find that parents underreport significantly more than head teachers. The reason for this is that parents choose to report on days when they know teachers are more likely to be present (reports are expected once a week). In terms of actually reducing absenteeism, there is only a significant effect from head teacher monitoring with financial incentives. A possible interpretation of this finding is that given the current equilibrium parents are simply uncomfortable holding teachers to account. This indicates that there are limits to how much can be achieved via the short route of accountability or local beneficiary control alone, in overcoming rent-seeking, especially when local institutions are weak and lack bite. In such cases, top-down monitoring may be a more effective way to reduce rent-seeking and corruption, especially in a context where 18

21 communities are mobilized to respond to adverse findings from such monitoring through social pressure. This is illustrated by an anti-corruption experiment evaluated by Olken (2007) in Indonesia in the context of the country s sub-district development program. Villages receive a grant and decide in local meetings how to spend the grant with the majority allocated to infrastructure projects. The funds and their uses are monitored in a series of village meetings in which the village implementation team presents the accounts and progress, though attendance at these meetings is usually very low. In addition, there is the possibility (about a 4 percent chance) of being audited by the independent government development audit agency. If these audits uncover corruption, they usually do not lead to criminal prosecution, but the results are reported to village meetings, which can result in social sanctions. Against this backdrop, Olken examines three randomized interventions. In the first, villages are told that their project will be audited by the government agency for sure. In the second and third, participation in the village meetings is encouraged through invitations, and invitations coupled with anonymous complaint forms. Several interesting findings emerge. First, the author finds that the top-down audit reduces corruption by 8 percentage points (1/3) as measured by the percentage of missing project expenditure. Second, the invitations intervention had no effect on the overall level of missing expenditure. Finally, the anonymous complaints forms reduced missing expenditure, but only when distributed through village schools rather than the village councils. This indicates that without credible bottom-up monitoring, increasing local control may not constitute a credible tool to monitor corruption. That is not to say though that local involvement does not matter but that it may not suffice. Indeed, even the first treatment arm contains a large bottom-up component. As noted in Fox (2015): The causal mechanism behind the audits rarely involved official penalties. It was mainly the threat of community responses to the promised local dissemination that gave the audits the clout to reduce corruption. 19

22 National Reforms Local reforms may also show their limits when there are attempts to scale them up. While short-term contracts for teachers may be a promising method to increase local accountability of teachers, large-scale reforms of this type may threaten established rent-sharing relationships. In 2009, the Kenyan government decided to roll out a nationwide program in which 18,000 teachers were hired on short-term contracts. Bold et al. (2016) study the pilot-phase of this program and compare government management of contract teachers against a benchmark of NGO management. Evaluating the experimental implementation against the backdrop of the nationwide roll-out, the authors are able to show evidence of how the reform was partly undermined because of the power balance between government and teacher unions. They find positive and significant effects of the program only in schools where the contract teacher program was administered by an international NGO. Placing an additional contract teacher in a school where the program is managed by the NGO increased test scores by roughly 0.18 standard deviation. When moving from NGO to government implementation, the effects of the program were virtually undone: treatment effects were significantly smaller and indistinguishable from zero in schools receiving contract teachers from the Ministry of Education. The authors show that there is evidence both of capacity constraints and power constraints when the government implemented a contract teacher program: schools in which the contract teacher was managed by the government received fewer monitoring visits and experienced significant salary delays. But there were also important political economy reasons for the program s failure: the government s plan to employ 18,000 contract teachers nationwide posed a significant threat to the Kenyan National Union of Teachers. The union waged an intense political and legal battle against the contract teacher program. The political battle changed the program in important ways and questioned the government s commitment to main incentive mechanism, namely that tenure was conditional on performance. After two years, the government gave in and employed all the 18,000 teachers (though not those in the experiment) permanently. Bold et al. find that this political battle affected the contract teachers in the experimental government treatment arm, who identified more closely with the union and their demands. 20